GPT-6.1 Terra
Points and comments are a snapshot, not live.
GPT-6.1 Terra matches GPT-6 Astra's benchmark performance at one-tenth the price per token.
OpenAI released GPT-6.1 Terra, a model focused on visual reasoning, computer use, and cost efficiency. It scores 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam at medium reasoning effort, at about one-third the cost. On MMMU-Pro, it nearly matches GPT-6 Astra and GPT-6.1 Sol. Its hallucination error rate dropped from 53.6% to 16.7% at low reasoning. On Terminal-Bench 4.0, it solves 50.3% of tasks at max effort, more than double GPT-5.6 Terra's 21.5%. Standard API pricing is $1 per million input tokens and $5 per million output tokens.
What commenters are saying
Commenters are split on how representative OpenAI's benchmark selection is. Some see GPT-6.1 Terra as effectively matching GPT-6 Astra at a fraction of the cost, while others point to Terminal-Bench 4.0 results where Terra at high reasoning costs more than GPT-6.1 Sol for worse performance. Several commenters note the release targets enterprise computer-use and automation workflows, not pure coding. A few criticize OpenAI's "cost per task" framing as weaselly, arguing it masks recent price increases for subscribers. The thread also discusses whether OpenAI can compete with Anthropic's Opus 5.5 on software engineering tasks.